JP2576079B2 - Character extraction method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an optical character reader.
(Hereinafter also referred to as “OCR”).

[0002]

2. Description of the Related Art In extracting a character pattern from a character string, a circumscribed rectangular coordinate of a black pixel connected component is obtained, and an integrated pattern in which a result of integrating a plurality of circumscribed rectangles is equal to a character size estimated in advance is cut. The method of issuing is common. However, in particular, in the case of a free handwritten character string, contact between a plurality of characters and character parts occurs, so that it is necessary to forcibly cut the circumscribed rectangle based on the detection result of the character boundary. As one of the methods, there is a method using a line density obtained by counting the number of character lines in a direction perpendicular to the direction of the character string. In this method, the rectangle is divided near the center of the rectangle to be divided where the linear density in the vertical direction becomes minimum. Regarding this, for example, Nakabayashi et al., “Fast-frame-free handwritten character string reading method using fuzzy search” (IEICE (D-II), J.
74-D-II, 11, PP.1528-1537) is helpful.

[0003]

However, the above-mentioned prior art focuses on only local information of the character shape, and the character string period in which the character entry pitches in the character string tend to be equal. Sex is not specifically considered. Further, when a plurality of characters or parts of the characters are in contact with each other as shown in FIG. 3, the search range of the line density is not always near the center. There was a problem that it could not be put out. The present invention has been made in view of the above circumstances, and an object of the present invention is to solve the above-described problem in the conventional technology and to consider the periodicity of a character string so that a plurality of characters or character portions may be contacted. It is an object of the present invention to provide a character extracting method capable of accurately extracting a character from a free handwritten character string written in the form.

[0004]

SUMMARY OF THE INVENTION The above objects of the present invention are as follows.
Obtain image data by optically scanning character strings
And the acquired image data in the direction of the character string.
Scan in the orthogonal direction to find the coordinates of the circumscribed rectangle of the black connected component
The circumscribed rectangle calculation processing step to be sought and the step
The number of black pixels in the direction orthogonal to the direction of the character string.
Marginal distribution calculation processing step, and the circumscribed rectangle is a character string
File to determine whether to integrate in the direction orthogonal to the
Last merge processing step and after the first merge
Character size estimation processing for estimating the character size from a rectangle
Step and sentence estimated to be the length of the bounding rectangle in the string direction
Determines whether to perform forced cutting based on the character size ratio
Forced disconnection determination processing step
Divide the circumscribed rectangle as necessary based on the fixed result
The forced cutting process step and the rectangular coordinates are integrated in the direction of the character string.
A second merge processing step to determine whether
In the character cutting method,
Is forcibly cut by the forcible cut determination processing step.
Marginal distribution within the circumscribed rectangular area
A smoothing processing step for smoothing at least two pitches
And smoothing at the different pitches by the step described above
Comparing the processing results to find the range to search for the forced cutting point
And a forced disconnection search range detection processing step
Forcing detected by the forced disconnection search range detection processing step
Find the position where the marginal distribution is minimum from the cutting search range
This is achieved by a character segmentation method characterized by dividing a circumscribed rectangle at a lever position .

[0005]

In the character extracting method according to the present invention, attention is paid to the result of smoothing (moving averaging) the peripheral distribution of characters at different pitches. That is, in the OCR, pixel data obtained by raster-scanning a form or the like is processed, and the peripheral distribution is a value obtained by counting the number of black pixels in a direction perpendicular to the direction of a character string in the data. Characters have characteristics such as the tendency that the number of black pixels near the center tends to be large, and the pitch at which characters are written tend to be equal within a character string. Can be caught. FIG. 2 shows an example of a character string. In FIG. 2, 21 is the original pattern of the character, 22 is the acquisition result of the peripheral distribution,
Reference numeral 23 denotes the number of black pixels after performing a smoothing process twice on the marginal distribution with a pitch of 1/4 of the character size estimated in advance. The pitch indicates the number of black pixels after performing the smoothing process . As is clear from the numbers of black pixels 23 and 24 in FIG. 2, the number of black pixels after the smoothing process is such that the local fluctuation is removed and the periodic fluctuation tendency of the character string is reflected. . When the variation tendency of the number of black pixels is examined in more detail, a character peak appears remarkably for each character in the number of black pixels 23 in FIG. In addition, there is a character boundary between the intersection 24 of the peripheral distribution acquisition result 24 after detecting the peak of the black pixel number 23 in FIG. 2 and the black pixel number 23, and the boundary is the black pixel number 23. It can be understood that the position coincides with the position where the minimum value is obtained.

[0006]

Embodiments of the present invention will be described below in detail with reference to the drawings. FIG. 1 is a block diagram showing a configuration of a character extracting device according to one embodiment of the present invention. Image data read from an image input device such as a scanner is stored in the image memory 1 in FIG. The black connected component circumscribed rectangle calculation circuit 2 scans a character string in the image memory 1 in a direction orthogonal to the direction of the character string to obtain a black connected component circumscribed rectangle. FIG. 3A shows an example of the obtained result. In parallel with this, the margin distribution calculation circuit 3 calculates the margin distribution by counting the number of black pixels in a direction orthogonal to the direction of the character string. An example of the results calculated et al was shown in 22 of FIG. Next, the first merge circuit 4 performs integration in a direction orthogonal to the direction of the character string of the black connected component circumscribed rectangle.
That is, if the length of the overlapping portion of the mapping to the coordinate axis parallel to the direction of the character string is longer than １ ／ of the shorter value of the lengths of the overlapping circumscribed rectangles in the character string direction, Perform integration. An example of the circumscribed rectangle after the integration is shown in FIG.
Note that this first merge processing is described in the above sentence.
Contribution (Religion (D-IID, as described in
This is a normal process for a trader.

As described above, since a character has a characteristic that the width and height of the character are substantially equal, if either the width or the height of the character can be estimated, this value can be considered as the character size. . The character size estimating circuit 5 estimates the character size by obtaining the average or median of the lengths in the direction orthogonal to the direction of the circumscribed rectangular character string after the first merge, based on such characteristics of the character. .
Next, the forced disconnection judging circuit 6 calculates the ratio between the length of the circumscribed rectangle in the character string direction after the first merge and the character size estimated value, and when this value is equal to or larger than the threshold value,
The circumscribed rectangle is set as a forced disconnection candidate. If the threshold value is 1.2, 31 in FIG. 3B is a forced disconnection candidate. The smoothing circuit 7 performs smoothing of the marginal distribution in the region (31 in FIG. 3B) that has become the forced cutting candidate. The smoothing is based on a moving average method for each pixel . Here, 1/4 of the character size is used as the pitch and 2
Times smoothing was performed . An example of the obtained result is shown in FIG.
23 (actually, only the portion surrounded by a broken line) , and the result of further smoothing the result by using the character width as the pitch is shown in FIG. In this way, the region that became a forced disconnection candidate
Performs marginal distribution smoothing only within the area (31).
As a result, it is possible to suppress unnecessary increase in the processing amount.
You.

Next, in the forced cutting search range detecting circuit 8,
By using the above-described smoothing result, the forced cutting search range is set as shown in FIG.
Is determined by the method shown in FIG. This is 0005
Between the intersection of 24 and 23 in FIG. 2 and the next intersection
This is equivalent to the process of detecting the boundaries of existing characters. Step 41: Initial Setting The tracking start position of the smoothing result is registered. Step 42: clear flag flg_one (a flag indicating that a ridge has been detected),
flg_tani (flag indicating that a valley has been detected),
The flg_start (flag indicating the start point of the cutting search range) and the flg_end (the flag indicating the end point of the cutting search range) are turned off. Step 43: Acquisition of Smoothing Result at Tracking Position Using 1/4 of the character size as a pitch, abrasion results obtained by smoothing twice are set as a, and a as a result of further smoothing using a character size as a pitch is set as b. Step 44: a-b turns on the flg_one larger than a threshold value.

If flg_start and flg_end are on, the registered search start and end points are output, and flg_start and flg_end are output.
Turn off start, flg_end, and flg_tani. Step 45: flg_one is on and flg_start required off,
In the case of a ≦ b, the current tracking position is registered as a search start point. flg_star
Turn on t. Step 46: When flg_start is on and ba is larger than the threshold value flg_tani is turned on. Step 47: flg_tani is on, flg_end is off,
If a ≧ b, the tracking position of the current point is registered as the search end point. flg_en
Turn d on. Turn on flg_tani. Step 48: If the tracking position is within the target area, proceed to step 43; otherwise, end.

Next, the forcible disconnection circuit 9 forcibly disconnects
In the forced disconnection search range obtained by the search range detection circuit 8,
Forcibly cut the circumscribed rectangle at the position where the marginal distribution is minimal . An example of the obtained result is shown in FIG. The second merge circuit 10 attempts integration in the direction of the character string of the circumscribed rectangle, and obtains a circumscribed rectangle so that the aspect ratio of the circumscribed rectangle after integration is close to 1. An example of the result obtained is
It is shown in FIG. This is the final cutout result. Note that this second merge processing is also described earlier.
Document (IEICE (D-II)).
This is a normal process for a trader. According to the above embodiment,
Even when extracting a character from a free handwritten character string in which contact between characters or parts of the character occurs, it is possible to obtain an effect that a sentence pattern can be accurately extracted. In FIG.
The configuration of an OCR to which the character cutout device shown in the above embodiment is applied is shown. In the figure, 51 is an image input device such as a scanner,
Reference numeral 52 denotes a character cutout device according to the present invention, 53 denotes a character feature extraction unit, 54 denotes a character identification unit, and 55 denotes a display unit of an identification result.

The effect of the present apparatus is that the character can be recognized with high accuracy as a result of performing the character recognition based on the effect of the character extracting apparatus described above. Note that the above embodiment is an example of the present invention, and it goes without saying that the present invention should not be limited to this. For example, the smoothing process 7 shown in FIG. 1 does not necessarily need to be performed on all character strings, but may be performed only on characters or portions where contact occurs between characters.

[0012]

As described above in detail, according to the present invention, by taking into account the periodicity of a character string, a character can be accurately extracted from a free handwritten character string written by touching a plurality of characters or character parts. This has a remarkable effect that a character extracting method capable of extracting characters can be realized.

[0013]

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a character extracting device according to an embodiment of the present invention.

FIG. 2 is a diagram illustrating a smoothing result of a marginal distribution and a front-side distribution.

FIG. 3 is a diagram illustrating a flow of processing of the character cutout device according to the embodiment using a black connected circumscribed rectangle.

FIG. 4 is a diagram showing a processing flow of a forced disconnection search range detection method.

FIG. 5 is an OCR to which the character extracting method according to the present invention is applied.
FIG. 3 is a block diagram of the configuration of FIG.

[Explanation of symbols]

1: image memory, 2: black connected component circumscribed square calculation circuit,
3: marginal distribution calculation circuit, 4: first merge circuit,
5: character size estimation circuit, 6: forced disconnection determination circuit, 7:
Smoothing circuit, 8: forced cutting search range detection circuit, 9: forced cutting circuit, 10: second merge circuit, 21: original pattern of character, 22: acquisition result of peripheral distribution, 23: result of smoothing processing of peripheral distribution , 24: The number of black pixels as a result of further smoothing the number of black pixels 23.

Claims (1)

(57) [Claims] An image is obtained by optically scanning a character string.
Acquiring the data and writing the acquired image data
Scan in the direction perpendicular to the direction of the character string to circumscribe the black connected component
A circumscribed rectangle calculation processing step for obtaining coordinates of the rectangle;
In parallel with the step, black screen in the direction orthogonal to the direction of the character string
A marginal distribution calculation processing step of counting prime numbers;
Whether to combine rectangles in a direction orthogonal to the direction of the string
A first merge processing step for determining
Character size that estimates the character size from the circumscribed rectangle after merging
Size estimation step and the length of the circumscribed rectangle in the character string direction.
Force disconnection from the estimated character size ratio
A forced disconnection determination processing step for determining whether or not
If necessary, based on the result of the judgment by PUB,
Forced cutting step to split the shape
Second merge processing to determine whether to merge in the column direction
A character extracting method having a logical step.
The disconnection processing step is replaced with the forced disconnection determination processing step.
Peripheral area within the circumscribed rectangular area that was subject to more forced cutting
Smoothing to smooth the fabric result with at least two pitches
Processing step and the different pick
Search for the forced cutting point by comparing the results of the smoothing process
In the forced cutting search range detection processing step
And the forced disconnection search range detection processing step
The marginal distribution is minimized from the detected forced cut search range.
And extracting a circumscribed rectangle at this position . "